task_manager_agent/README.md

Task Manager Agent — Personal AI OS

Automatically turns emails and meeting notes into prioritized, synced tasks.

What it does

Capability	Detail
Email → Tasks	Scans inbox every 15 min, extracts actionable items via Groq
Meeting → Tasks	Parses Google Calendar descriptions for action items and follow-ups
Prioritization	LLM scores each task 1–10 using urgency + impact + effort
Deduplication	Never creates the same task twice
Sync	Pushes to Notion DB and/or Todoist
Digest	Sends a formatted email + optional WhatsApp summary
Triggers	New email detection (polling) + daily 9 AM scheduled sync

---

File structure

task_manager_agent/
├── main_agent.py       # Orchestrator, scheduler, email watcher
├── data_fetcher.py     # Gmail + Calendar + existing task fetch
├── llm.py              # Groq extraction + prioritization prompts
├── task_store.py       # Local JSON persistence + dedup
├── delivery.py         # Email SMTP, WhatsApp, Notion, Todoist
├── .env.example        # All env vars with explanations
├── requirements.txt
└── README.md

---

Setup

1. Install dependencies

pip install -r requirements.txt

2. Configure environment

cp .env.example .env
# Edit .env with your credentials

3. Google credentials

Share the same credentials.json and token.json from your earlier agents. The agent needs scopes:

• gmail.readonly
• calendar.readonly

4. Notion Database setup

Create a Notion database with these properties:

Property	Type
Name	Title
Status	Select: To Do, In Progress, Done
Priority	Select: Critical, High, Medium, Low
Due Date	Date
Category	Select: Work, Personal, Admin, Communication, Research, Finance, Health, Other
Source	Rich Text
Priority Score	Number
Notes	Rich Text

Copy the DB ID from the Notion URL: https://notion.so/your-workspace/THIS-IS-YOUR-DB-ID?v=...

5. Run

python main_agent.py

On startup the agent runs immediately, then enters the continuous loop:

• Email watcher polls every EMAIL_POLL_INTERVAL_MINUTES minutes
• Full sync fires daily at 09:00

---

How prioritization works

The LLM scores each task using four axes:

Priority Score (1-10) = f(urgency, impact, effort, dependencies)

Score	Label	Example
9–10	Critical	"Contract due tomorrow, client blocked"
7–8	High	"Respond to investor by Friday"
5–6	Medium	"Update project docs"
1–4	Low	"Read that article someone forwarded"

---

Cron alternative

To run via system cron instead of the built-in scheduler:

# Daily 9 AM sync
0 9 * * * cd /path/to/task_manager_agent && python -c "from main_agent import run_task_extraction_pipeline; run_task_extraction_pipeline('cron_9am')"

# Email polling every 15 min
*/15 * * * * cd /path/to/task_manager_agent && python -c "from main_agent import run_task_extraction_pipeline; run_task_extraction_pipeline('email_poll')"

---

Agent position in Personal AI OS

01 ✅ Daily Planner Agent
02 ✅ Email Agent
03 ✅ Meeting Prep Agent
04 ✅ End-of-Day Review Agent
05 ✅ Task Manager Agent  ← YOU ARE HERE
06    Research Agent
07    Finance Agent
08    LinkedIn Agent
09    Knowledge Agent
10    Master Orchestrator (Mem0 + LangGraph)

The Master Orchestrator will call run_task_extraction_pipeline() directly and read from TaskStore to feed task context into other agents.